Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What Is AGI-Hard

lspace.swyx.io

Post reply on HN
Print view View original

71–80 of 129 posts

Re: What Is AGI-Hard

#71

Post by cameronh90 »

fartsucker69 wrote: post #49
albertzeyer wrote: post #42

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this leve…

It's totally general in the sense that we can logically approach arbitrary problem types and there is nothing known that we can't approach with our tools of intelligence, so to speak. That doesn't necessarily mean that we are actually good at solving those problems or that there isn't an upper level of difficulty that we will just not be able to solve practically, but we are definitely able to approach all classes of…

I've long thought there is some sort of intelligence parallel to Turing Completeness: given enough time, resources and experimentation, it can figure anything out.

At least at a super-organism level, humanity has achieved that, in my opinion. However, I'm not sure whether any individual human can be said to meet that criteria. Perhaps some exceptional examples meet that criteria, or maybe there is some vague IQ cut-off where any human could meet that criteria if given a really long time to study a problem - but I'm fairly certain that many humans don't meet that criteria individually.

Obviously this isn't a mathematical or scientifically provable statement, just a general concept that gets a bit fuzzy at the edges.

What I would add is a lot of people seem to judge AI overly harshly compared to humans, perhaps comparing them against humanity as a super-organism or at least an above average human. Sure, ChatGPT makes weird mistakes and can't write code as well as me, but compared to many people I know, it's a far better programmer and possibly less error prone in general. Too many people are like "ha it can't solve this differential equation", but my mum doesn't even know what a differential equation is, and thought that the moon was the sun at night. At least ChatGPT knows the moon is a distinct celestial body to the sun.

Re: What Is AGI-Hard

#72

Post by dbspin »

fartsucker69 wrote: post #69
dbspin wrote: post #55

Earlier quoted context omitted.

> there is nothing known that we can't approach with our tools of intelligence, so to speak. Such things may necessarily be unknowable. If there are classes of observations that we are unable to make, or problems that we are unable to comprehend, its absolutely plausible that they would be cognitive blindspots. In fact there's a whole theoretical strand in cognitive science / ontology that suggests this. We perceive…

even hypothetically this is only partially possible. for sure what you talk about would be something that is purely an abstract / logical problem, because if something we effectively "can't think" originates somehow in the real world, we would be able to observe it and realize that we can't reason about it. and even when we have exhausted everything there is to know in the universe and solved every type of problem, w…

This has already been replied to but no, this is not a 'God of the gaps' argument. As I mentioned it's a thread in ongoing research on cognition. There's a good summary of some of the perspectives in this space here - https://www.quantamagazine.org/the-evolutionary-argument-aga...

Re: What Is AGI-Hard

#73

Post by dbspin »

jvm___ wrote: post #63
dbspin wrote: post #55

Earlier quoted context omitted.

> there is nothing known that we can't approach with our tools of intelligence, so to speak. Such things may necessarily be unknowable. If there are classes of observations that we are unable to make, or problems that we are unable to comprehend, its absolutely plausible that they would be cognitive blindspots. In fact there's a whole theoretical strand in cognitive science / ontology that suggests this. We perceive…

What's your definition of fitness in this case? Survival of the fittest?

Here's a background on some of the arguments - https://www.quantamagazine.org/the-evolutionary-argument-aga...

Re: What Is AGI-Hard

#74

Post by snordgren »

albertzeyer wrote: post #59
TacticalCoder wrote: post #50

Earlier quoted context omitted.

> What does human-level intelligence mean? What about a simple test: it means being able to create better versions of ChatGPT and StableDiffusion. Can an AI create a better version of itself, all by itself? Because humans are going to create better versions of ChatGPT. So no matter the exact definition of "intelligence" or "human intelligence", the goal is simple: create a better version of itself and it'll be consid…

So, most humans are not intelligent then? Because I don't think that an average human can create a better version of ChatGPT / StableDiffusion. Or a better version of itself, if that is the baseline. Or do you count evolution here? But evolution does not necessarily lead to higher intelligence. But also: Once you have some model or algorithm for intelligence, where you just need to scale up the size / training data,…

We already know that humans are human-level intelligent, so we don't need to devise a test to prove that they are.

If AI >= highest level of human intelligence, then AI > human intelligence. It is less accurate because it will miss any AI that achieves a precisely human level of intelligence, but it is probably also an easier test to define than others.

Of course simply increasing the scale of the model does not count as an improvement, there is nothing novel in that suggestion. But you are right that it opens a can of worms as to what an improvement is. Does an improvement count if it improves the model in the 80% of scenarios that only appear 20% of the time, while similarly reducing its performance in the 20% of scenarios that appear 80% of the time?

Re: What Is AGI-Hard

#75

Post by adrian_b »

dbspin wrote: post #52
TacticalCoder wrote: post #50

Earlier quoted context omitted.

> What does human-level intelligence mean? What about a simple test: it means being able to create better versions of ChatGPT and StableDiffusion. Can an AI create a better version of itself, all by itself? Because humans are going to create better versions of ChatGPT. So no matter the exact definition of "intelligence" or "human intelligence", the goal is simple: create a better version of itself and it'll be consid…

> What about a simple test: it means being able to create better versions of ChatGPT and StableDiffusion. This doesn't follow at all. An ant will haphazardly create a better (fitter) version of an ant simply through breeding and natural selection. Meanwhile humans cannot intentionally create 'better versions' of humans (with any degree of reliability for a given selection criteria). There's no intelligence on earth (…

While it is impossible to predict if and when humans will become able to make other more intelligent humans, there are good chances that during this century, maybe already during the first half of this century, it will become possible to make versions of humans with various simple improvements, for instance with the ability to make vitamin C in the liver, like all our more distant ancestors and relatives, in order to no longer be dependent on an adequate daily intake from external sources.

Re: What Is AGI-Hard

#76

Post by miguelrochefort »

sanxiyn wrote: post #8

> Can AI invent calculus from first principles? The thing is, most people can't. In fact, if you reinvented calculus yourselves, you should be very proud of your mathematical ability.

I would be surprised if we did not already have AIs that could invent calculus from first principles.

LLMs solve natural langage and common sense, which are not necessary for discovering calculus.

Re: What Is AGI-Hard

#77

Post by mwigdahl »

comboy wrote: post #64
bmitc wrote: post #58

Earlier quoted context omitted.

> As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. That’s kind of a silly semantic argument though. It also assumes that everyone interested in AGI is trying to make “human-leve…

> So I think AGI is fine unless you’re explicitly trying to mimic human intelligence. Humans we will only recognize AGI when it can simulate all our flaws. We won't think the computer is able to see properly until it can observe the same optical illusions that we do (which are visual system flaws). By general we mean being able to completely relate to *our* experience. We are very biased even when it comes to say mat…

I think this point is often missed. We live in daily contact with pet animals that think entirely differently than us, and even humans with special needs behave in radically different ways and value radically different things in their lives. If tiny variations in the human brain are able to wreak such profound differences in motivation, behavior, and cognition, why are people fixated on expecting banal human hormonally-driven power-seeking and survival behavior from a constructed mind with a totally different architecture in an entirely different substrate?

Re: What Is AGI-Hard

#78

Post by mannykannot »

janalsncm wrote: post #16
Animats wrote: post #9

Earlier quoted context omitted.

> This article is a bit too hand-wavy for my taste. From the article: "I think this is the sort of debate that makes good podcast fodder". So the article author admits that. It's hand-wavy on vision. Parts of vision are not "AGI-hard". A key component of self-driving is something that can determine whether a volume of space is empty, occupied, or ambiguous. It's necessary to have something that does not report "empty…

Very well said. Just want to add one thing. > This is a big problem now that there are systems able to produce huge volumes of plausible blithering that's factually wrong. If because of some law of nature ChatGPT were the last iteration of automatic text generation we could ever create, it would be a big and permanent problem. Right now it’s just a big and temporary problem, with a huge opportunity for automatic fact…

What you call a huge opportunity is actually (or also, if you prefer) a major shortfall in the current LLMs. Whether it will turn out to be merely temporary, and more pertinently, how it might be solved and whether it can be solved with more of the same, are themselves unanswered questions at this time.

Re: What Is AGI-Hard

#79

Post by ipunchghosts »

TacticalCoder wrote: post #50
albertzeyer wrote: post #42

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this leve…

> What does human-level intelligence mean? What about a simple test: it means being able to create better versions of ChatGPT and StableDiffusion. Can an AI create a better version of itself, all by itself? Because humans are going to create better versions of ChatGPT. So no matter the exact definition of "intelligence" or "human intelligence", the goal is simple: create a better version of itself and it'll be consid…

Isnt this what a Gödel machine is supposed to do.

Re: What Is AGI-Hard

#80

Post by mannykannot »

bmitc wrote: post #60
Retric wrote: post #57

Earlier quoted context omitted.

Humanity has developed technology to improve humanity. Glasses correct issues, and microscopes added new capacity. We haven’t gone down the genetic engineering path but we could fix issues with our DNA.

> Humanity has developed technology to improve humanity. I would argue that the vast majority of technology has hurt humanity and that a large proportion of technology is simply solving problems caused by other technology.

I am not entirely convinced you really hold this view, but if so, you could reduce technology's pernicious effects on yourself by not participating in this discussion.
Post reply on HN